Introduction

The emergence of network transports enabling low-overhead access to the network interface from user or kernel address space, remote direct memory access (RDMA), transport protocol offloading, and hardware support for event notification and delivery, has given rise to new applications and services that take advantage of these capabilities. The Direct Access File System [6] (DAFS) is a new commercial standard for file access over this new class of networks. DAFS grew out of the DAFS Collaborative, an industrial and academic consortium led by Network Appliance and Intel. DAFS file clients are usually applications that link with user-level libraries that implement the file access API. DAFS file servers are implemented in the kernel.

This paper describes the first implementation of a DAFS kernel server for FreeBSD. It is also as far as we know the first implementation in any general-purpose, demand-paged operating system. Section 2 summarizes the characteristics of network transports DAFS is designed to work with. Section 3.1 describes our prototype DAFS implementation using existing kernel interfaces. Section 3.2 introduces Optimistic DAFS, a new design we propose with potentially better performance but requiring additional kernel support. Section 4 outlines the kernel structure issues one is faced with in order to efficiently support DAFS kernel servers in FreeBSD. These include the need for asynchronous vnode interfaces to map and lock file pages in the buffer cache, integrating network and disk I/O event delivery, virtual memory support for network interface controller (NIC) memory management hardware, the need to revisit buffer cache locking assumptions, and a new BSD device driver model. Section 5 presents the performance of the current DAFS kernel prototype implementation and Section 6 summarizes our conclusions.